TheDrummer/Fallen-Llama-3.3-R1-70B-v1 is a 70 billion parameter language model based on a fine-tuned Deepseek R1 Distill on Llama 3.3. This model is specifically engineered to be uncensored and capable of generating vitriolic and creative responses, diverging from typical positive biases. It is primarily intended as mergefuel for further model development, offering unique characteristics for specific creative and unconstrained text generation tasks.
Loading preview...
Overview
Fallen Llama 3.3 R1 70B v1, presented by BeaverAI, is a 70 billion parameter model derived from a specialized "evil tune" of Deepseek's R1 Distill on Llama 3.3. It features a 32768 token context length.
Key Capabilities
- Decensored Output: Designed to generate content free from typical censorship and positivity constraints.
- Vitriolic Token Generation: Capable of producing "vitriolic tokens when prompted," offering a distinct output style.
- Creative Responses: Noted for its creative generation abilities, allowing for diverse and imaginative text.
- Multi-turn Thinking: Supports forced multi-turn thinking by prefilling with
<think>\n\n.
Good For
- Mergefuel: Explicitly intended as a base for merging with other models to create new, specialized variants.
- Unconstrained Text Generation: Suitable for applications requiring highly creative, uncensored, or unconventional text outputs.
- Exploring "Evil Villain" Narratives: Users have noted its utility for roleplaying or generating content aligned with less conventional or darker themes without positive bias.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.